How to read a data analysis you did not run yourself

You do not need to be a statistician to interview a results section. Five questions will surface most of the problems.

Project Tutor Research TeamResearch & Advisory · 16 June 2026 · 4 min read

What was measured, exactly

Ask for the definition of every variable in plain language. A surprising share of confusing results comes from a measure that does not mean what the reader assumes it means, such as an active user defined over a window that quietly changed mid-study.

Who is missing from the data

Response rates, dropouts and excluded records shape the answer more than the model does. Ask who was eligible, who responded, and how the excluded differ from the included.

How large is the effect, in units you care about

Statistical significance says an effect is unlikely to be noise. It says nothing about whether the effect matters. Insist on the effect size in real units: pounds, days, percentage points, customers.

What else could explain this

Ask what alternative explanations were considered and what was done about them. A results section that names its own competing explanations is usually more trustworthy than one that does not.

Share

Useful? Send it on, or ask us how it applies to your own work.

Ask about this

More Reading

All articles

Start today

Tell us about your research and we will take it from there.

+974 33246963